whisper-tiny-arabic-fine-tuned-v2
This model is a fine-tuned version of Mohammadawad1/whisper-tiny-arabic-finetuned-v1 on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.6705
- Wer: 48.6608
- Cer: 18.3751
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: constant_with_warmup
- lr_scheduler_warmup_steps: 50
- num_epochs: 4
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 0.1039 | 1.0 | 883 | 0.5912 | 47.6739 | 17.4026 |
| 0.076 | 2.0 | 1767 | 0.6078 | 47.8008 | 17.7679 |
| 0.0597 | 3.0 | 2650 | 0.6399 | 49.1874 | 18.6247 |
| 0.0374 | 4.0 | 3532 | 0.6705 | 48.6608 | 18.3751 |
Framework versions
- Transformers 4.28.0
- Pytorch 2.6.0+cu124
- Datasets 2.18.0
- Tokenizers 0.13.3